Single-nucleotide editing: from principle, optimization to application

Abstract

Cytosine base editors (CBEs) and adenine base editors (ABEs), which are generally composed of an engineered deaminase and a catalytically impaired CRISPR–Cas9 variant, are new favorite tools for single base substitution in cells and organisms. In this review, we summarize the principle of base editing systems and elaborate on the evolution of different platforms of CBEs and ABEs, including their deaminase, Cas9 variants, and editing outcomes. Moreover, we highlight their applications in mouse and human cells, and discuss challenges and prospects of base editors. The ABE- and CBE-systems have been used in gene silencing, pathogenic gene correction and functional genetic screening. Single-base editing is becoming a new promising genetic tool in biomedical research and gene therapy.

Introduction

Programmable nucleases, such as clustered regularly interspaced short palindromic repeats (CRISPR), zinc finger nucleases (ZFNs), and transcription activator-like effector nucleases (TALENs), have emerged as powerful tools for genome editing in various organisms (Doudna & Charpentier, 2014; Gaj, Gersbach, & Barbas, 2013; Sander & Joung, 2014). Collectively, genome editing with these programmable nucleases requires making DNA double-strand breaks (DSBs) and inducing cellular machinery, such as non-homologous end joining (NHEJ), or homology-directed repair (HDR), which repairs DSBs (Wiles, Qin, Cheng, & Wang, 2015). NHEJ generates insertions and deletion (indel) mutations, while HDR generates pre-defined precise modification. Unfortunately, HDR occurs frequently in the S- and G2-phases of a cell cycle, making it difficult to apply in postmitotic cells (Cox, Platt, & Zhang, 2015; Jeggo & Lobrich, 2007; Lin, Staahl, Alla, & Doudna, 2014; Ran et al., 2013). In addition, DSBs are preferentially repaired by the alternative pathway of NHEJ, which results in many non-targeted stochastic indels (Cong et al., 2013; Ran et al., 2013).

Recently, alternative genome editing strategies for precise base editing, such as cytosine base editors (CBEs) and adenine base editors (ABEs), were developed by several groups. CBEs and ABEs can directly convert one base pair to another, for instance A•T becomes G•C or G•C becomes A•T, at a target gene without reliance on HDR and introduction of DSBs (Gaudelli et al., 2017; Komor, Kim, Packer, Zuris, & Liu, 2016; Ma et al., 2016; Nishida et al., 2016). Generally, base editors function in one ribonucleoprotein with two activities: a catalytically impaired CRISPR–Cas9 variant targets a specific DNA sequence to generate a single-stranded DNA “bubble”, and it simultaneously deaminates its target, nicks off the non-targeted strand and catalyzes base conversion in this “bubble” (Gaudelli et al., 2017; Komor et al., 2016; Ma et al., 2016; Nishida et al., 2016).

The original studies of CBEs used different cytidine deaminases and domain permutations, and all designs led to an adequate efficiency in base editing. Their editing windows are approximately five base-pairs, depending on which fusion protein is used (Komor et al., 2016; Ma et al., 2016; Nishida et al., 2016). The original ABEs consisted of an engineered adenine deaminase and a CRISPR–Cas9 nickase. A high base editing frequency (~50% in human cells) was achieved through seven evolutions of ABEs (Gaudelli et al., 2017). Till now, base editors can achieve the direct programmable introduction of all four transition mutations (C to T, A to G, T to C, and G to A) without DSBs, which enables installing gene-correcting or gene-suppressing mutations in the animal, plant, microbe, and even human cells (Chen et al., 2017; Gu et al., 2018; Hua, Tao, Yuan, Wang, & Zhu, 2018; K. Kim et al., 2017; G. Li et al., 2017; Li, Sun, Du, Zhao, & Xia, 2017; Liang, Ding, et al., 2017; Liang, Sun, et al., 2017; Zhou et al., 2017).

In this review, we summarize the design and evolution of CBE and ABE platforms, including their deaminase, engineering of Cas9 variants, and editing outcomes. We highlight application of genome editing in mouse and human cells. Furthermore, we discussed the challenges and prospects of base editors. Base editors are becoming effective base editing tool for gene modification in various fields, including biomedical research and stem cell therapy.

1. Current base editing technology in genomic DNA

1.1. Creation and evolution of base editors of G•C to A•T in genomic DNA

Studies have shown that cytosine base editors (CBEs), which are generally composed of CRISPR-Cas9 variant, uracil glycosylase inhibitor (UGI) and a designed cytidine deaminase enzyme such as rat/human apolipoprotein B mRNA editing catalytic enzyme (APOBEC), lamprey PmCDA1 or human activation induced cytidine deaminase (AID), enable direct G•C to A•T substitution in the DNA sequence (Table 1) (Bohn et al., 2015; Conticello, 2008; Harris, Petersen-Mahrt, & Neuberger, 2002; Holland et al., 2018; Komor et al., 2016; Ma et al., 2016; Mukhopadhyay et al., 2002; Nik-Zainal et al., 2012; Nishida et al., 2016; Ouadani et al., 2016; Petersen-Mahrt & Neuberger, 2003; Stenglein, Burns, Li, Lengyel, & Harris, 2010; X. Wang et al., 2018; Yoshikawa et al., 2002; Zong et al., 2018). Without inducing DSBs or requiring a donor template, cytidine deaminase, guided by catalytically-dead Cas9 (dCas9), Cas9 nickase (Cas9n) or its variants, mediates the direct conversion of cytidine (C) to uracil (U), resulting in G•U mismatches. Accordingly, repair mechanisms will resolve these mismatches, by turning G•U base pairs into A•T base pairs. Thus, these base editors can be used to produce irreversible G•C to A•T substitution in genomic DNA (Komor et al., 2016) (Figure 1A).

Table 1:

Cytidine deaminase enzyme used in base editing.

Cytidine deaminase enzyme	Abbreviation		Source	Editing activity	Main function	Reference
Apolipoprotein B mRNA editing catalytic subunit 1	APOBEC1		Rat/human	RNA, DNA	RNA deaminase, mediates C to U editing of mammalian apolipoprotein B RNA; and DNA deaminase in bacteria and in vitro.	Mukhopadhyay et al., 2002 Harris, Petersen-Mahrt, & Neuberger, 2002 Petersen-Mahrt & Neuberger, 2003;
apolipoprotein B mRNA-editing enzyme, catalytic polypeptide-like 3	APOBEC3	APOBEC3A	Human	DNA	DNA deaminase, introduces G-to-A hypermutations; and acts in a defense pathway against retroviruses and mobile elements. Selectively targets single-stranded DNA and does not deaminate double-stranded DNA or single-or double-stranded RNA.	Conticello, 2008 Stenglein, Burns, Li, Lengyel, & Harris, 2010 Bohn et al., 2015 Nik-Zainal et al., 2012
		APOBEC3B	Human	DNA
		APOBEC3C	Human	DNA
		APOBEC3D	Human	DNA
		APOBEC3E	Human	DNA
		APOBEC3F	Human	DNA
		APOBEC3G	Human	DNA
		APOBEC3H	Human	DNA
Activation-induced deaminase	AID		Human	DNA	Single-stranded DNA-specific cytidine deaminase, which induces somatic hypermutation (SHM), gene conversion in B-lymphocytes through deaminating C to U during transcription of Ig-variable and Ig-switch region DNA	Yoshikawa et al., 2002 Ouadani et al., 2016
N/A	pmCDA1		lamprey	RNA, DNA	AID orthologue, mediates C to U in mRNA or DNA.	Holland et al., 2018

FIGURE1. Two types of base editing.

(A) Apolipoprotein B mRNA editing catalytic enzyme (APOBEC)-based cytosine base editor (CBE) mediates G•C to A•T base editing. Cas9 nikase (Cas9n, blue), which is guided by a single guide RNA (sgRNA, orange), targets specific cytosine (C, red), nicks the non-edited strand in genomic DNA, and mediates separation of local DNA strands. A tethered APOBEC enzyme (brown) acts on the C in the targeted DNA single-strand, and mediates it to uracil (U). A tethered uracil DNA alycosylase inhibitor (UGI, pink) can block the base excision to protect G•U intermediate. The resulting G•U heteroduplex can be permanently converted to an A•T base pair through DNA replication or DNA repair.

(B) Adenine base editor (ABE) mediated A•T to G•C base editing. Cas9n (blue), which is guided by a gRNA (orange), targets specific adenine (A, red), nicks the non-edited strand in genomic DNA, and mediates separation of local DNA strands. An engineered TadA* enzyme (purple) acts on the A in the DNA single-strand, and deaminates it to inosine (I). The resulting T•I heteroduplex can be permanently converted to a C•G base pair through DNA replication or DNA repair. PAM: protospacer-adjacent motif.

1.1.1. APOBEC-based DNA base editors

There are mainly two types of CBEs ─ APOBEC-based DNA base editors and AID-based DNA base editors, depending on what kind of cytidine deaminase enzyme is used (Komor et al., 2016; Ma et al., 2016; Nishida et al., 2016). Since the original APOBEC-based DNA base editor was first published, many improvements have been reported to enhance its efficiency and precision (Y. B. Kim et al., 2017; Komor et al., 2017; Rees et al., 2017; L. Wang et al., 2017; Zafra et al., 2018).

APOBEC-based DNA base editor is mainly composed of a rat or human APOBEC deaminase and a CRISPR–Cas9 nickase. Editing efficiency, genome-targeting scope and precision have been increased through four generations of evolution of CBEs, including deaminase engineering and Cas9 variants engineering (Figure 2A) (Y. B. Kim et al., 2017; Komor et al., 2016; Komor et al., 2017; Rees et al., 2017).

FIGURE2. Four generations of Apolipoprotein B mRNA editing catalytic enzyme (APOBEC)-based cytosine base editors (CBEs).

Rat APOBEC1 has the ability to deaminate cytosine (C) to uracil (U), which has the base-pairing property of thymine (T). Fusing rat apolipoprotein B mRNA editing catalytic subunit 1 (rAPOBEC1) to catalytically-dead Cas9 (dCas9) enables RNA-programmed base substitution in genomic DNA. The first-generation cytosine base editor (CBE1) can efficiently convert cytosine (C) to thymine (T) in vitro. Fused uracil glycosylase inhibitor (UGI) to CBE1 creates the second-generation base editor (CBE2). The third-generation base editors (CBE3s) with engineered Cas9-cytidine deaminase fusions can increase the efficiency, genome-targeting scope, precision and specificity. Moreover, the fourth-generation base editors (CBE4s) contain two copies of UGI.

PAM: protospacer-adjacent motif.

(1). Fusing rAPOBEC1 with dCas9 for base editing in vitro

Canonical CRISPR-Cas9 system localizes to a target DNA sequence and natively creates a double-strand DNA backbone cleavage at the locus specified by the guide RNA, which results in random indels at the site of DNA cleavage through NHEJ, markedly limiting its application to point mutations in the target locus (Sander & Joung, 2014). A programmable system that can directly convert one DNA base to another at a programmable target locus without inducing DSBs can circumvent this restriction (Komor et al., 2016). The dCas9 containing Asp10Ala and His840Ala mutations, which inactivates Cas9’s nuclease activity, retains its ability to bind DNA with a guide RNA and avoids cleaving the DNA backbone (Jinek et al., 2012; Komor et al., 2016). In principle, combination of dCas9 with an enzyme that mediates the direct single-base conversion enables RNA-programmed base editing in genomic DNA (Komor et al., 2016).

The rat apolipoprotein B mRNA editing catalytic subunit 1 (rAPOBEC1), functioning as a cytidine deaminase, has the ability to deaminate cytosine (C) to uracil (U), which has the base-pairing properties of thymine (T) (Conticello, 2008; Komor et al., 2016). Hence, fusing rAPOBEC1 to the N-terminus of dCas9 (not the C-terminus, because it preserves deaminase activity) enables RNA-programmed base substitution in genomic DNA (Komor et al., 2016). However, conversion efficiency and deamination window depend on the linker length of rAPOBEC1-dCas9. Among different linker lengths, the 16-residue XTEN linker between rAPOBEC1 and dCas9 offers the greatest conversion efficiency and efficient deamination with a window of approximately five nucleotides (Komor et al., 2016; Schellenberger et al., 2009). The rAPOBEC1-XTEN-dCas9 proteins, as the first-generation cytosine base editor (CBE1), have preferred sequence context in an order of TC≥ CC≥ AC> GC, with C as the target (Komor et al., 2016) (Figure 2A). In particular, the maximal editing efficiency is achieved when the target C is at or near the position 7 (Komor et al., 2016).

(2). Evolving CBE1 for base editing in vivo

Studies have shown that the editing efficiencies of CBE1 are dramatically decreased in mammalian cells. Base excision repair (BER) is a cell’s primary response to G•U mismatches, and is initiated by excision of U by uracil DNA glycosylase (UDG), with the most common outcome being reversion of the U•G pair to a C•G pair (Komor et al., 2016; Kunz, Saito, & Schar, 2009). In order to protect the edited G•U intermediate from excision by UDG, uracil DNA glycosylase inhibitor (UGI) can be fused to the intermediate to block human UDG activity. The APOBEC–XTEN–dCas9–UGI proteins, as the second-generation base editor (CBE2), can inhibit BER and convert C to T through DNA replication (Komor et al., 2016; Mol et al., 1995) (Figure 2A). Editing efficiencies in human cells by CBE2 are increased up to 20%, compared to CBE1 (Komor et al., 2016).

(3). Improving editing efficiency of CBE2 with Cas9 nickase

A maximal base editing yield can be reached at 50% when converting and protecting the substrate strand of a C•G base pair. To exceed this limit, the catalytic His residue at the position 840 has to be restored in dCas9 of CBE2 to become Cas9n, which results in the third-generation base editor (CBE3, APOBEC1–XTEN–Cas9n–UGI) (Figure 2A) (Komor et al., 2016). CBE3 nicks the non-edited strand and keeps the integrity of the edited strand, and improves editing efficiencies up to 37% (Jinek et al., 2012; Komor et al., 2016; Pluciennik et al., 2010).

(4). Optimizing CBE3 with engineered Cas9-cytidine deaminase fusions

Soon afterwards, CBE3 was developed to expand the editing scope by using natural and engineered Cas9 variants with different protospacer-adjacent motif (PAM) specificities (Figure 2) (Y. B. Kim et al., 2017; Kleinstiver, Prew, Tsai, Nguyen, et al., 2015; Kleinstiver, Prew, Tsai, Topkar, et al., 2015). Base editors were further engineered to contain mutated cytidine deaminase domains that narrow the width of the editing window from ~5 nucleotides to as little as 1–2 nucleotides (Y. B. Kim et al., 2017). It has been noted that human APOBEC3A-cas9 fusion can effectively induce efficient base editing in both methylated DNA regions and GpC dinucleotides, thus expanding the scope of base editing (X. Wang et al., 2018; Zong et al., 2018). In addition, human APOBEC3A-Cas9n-UGI and APOBEC3B-Cas9n-UGI base editing complexes are more efficient than the original rat APOBEC1-Cas9n-UGI construct (St Martin et al., 2018).

Cas9 high-fidelity variant (Cas9-HF), which contains specific point mutations, is thought to have less binding energy with DNA than wild type Cas9 (wtCas9). The mutations presumably disrupt hydrogen bonding with the phosphate backbone of the complementary DNA strand, thereby decreasing Cas9 binding with mismatched sequences and increasing its overall specificity (Kleinstiver et al., 2016; Kleinstiver, Prew, Tsai, Topkar, et al., 2015; Liang, Sun, et al., 2017). Substitution of Cas9n with Cas9-HF in CBE generates high-fidelity base editor (HF2-BE2 and Cas9-HF1), and results in a substantially enhanced base editing specificity and efficiency, compared to CBE3 (Table 2) (Kleinstiver et al., 2016; Liang, Sun, et al., 2017). Moreover, some researchers have changed the delivery method of base editors and demonstrated that protein delivery of base editors maintains on-target base-editing efficiency and greatly enhances editing specificity, compared to previous plasmid transfections (Rees et al., 2017).

Table 2:

List of major base editors with their approximate editing windows and efficiency

Base editor	Construction (N to C)	PAM	Editing window	Efficiency	References
SaBE3	rAPOCBEC1-SaCas9n-UGI	NNGRRT (R=G or A)	N9 to N15	~50%	Y. B. Kim et al., 2017
CBE4	rAPOCBEC1-Cas9n-UGI-UGI	NGG	N13 to N17	~50%	Komor et al., 2017
SaBE4-Gam	Gam-rAPOCBEC1-SaCas9n-UGI-UGI	NNGRRT	N9 to N15	~50%	Komor et al., 2017
HF2-BE2	APOCBEC1-HF2-dCas9-UGI	NGG	N/A	100%	Liang, Sun, et al., 2017
BE-PLUS	nCas9-GCN4, scFv-APOBEC-UGI	NGG	N5 to N17	42.2%	Jiang et al., 2018
dCpf1-BEs	+/−NLS, rAPOCBEC1-YE-dCpf1-UGI, +/−3xFree UGI	TTTV	N8 to N13	~10–31%	Li et al., 2018
TAM	dcas9-AID (AIDx), +/−UGI	NGG	N12 to N16	52%	Ma et al., 2016
CRISPR-X	dCas9-directed MS2–AID*Δ	NGG	N50 to -N50	> 20%	Hess et al., 2016
Target-AID	dCas9-PmCDA1-UGI/Cas9n-PmCDA1-UGI	NG	N15 to N19	15%−55%	Nishida et al., 2016
TALE-AID ZFN-AID	TALE-AID ZFN-AID	N/A	N/A	2.5%	Luhan Yang et al., 2016
ABE7.10	TadA-TadA(7.10)*–XTEN–Cas9n–NLS	NGG	N13 to N17	58%±4.0%	Gaudelli et al., 2017
xABE	TadA-TadA(7.10)*–XTEN–xCas9–NLS	NG, GAA, and GAT	N13 to N17	N/A	Hu et al., 2018
SaABE	TadA-TadA(7.10)*–XTEN–SaCas9n–NLS	NNGRRT	N9 to N15	N/A	Hua, Tao, & Zhu, 2019
VQR-ABE	TadA-TadA(7.10)*–XTEN–VQR-Cas9n–NLS	NGA	N13 to N17	N/A	Hua, Tao, & Zhu, 2019 L. Yang et al., 2018
VRER-ABE	TadA-TadA(7.10)*–XTEN–VRER-Cas9n–NLS	NGCG	N13 to N17	N/A	Hua, Tao, & Zhu, 2019
Sa(KKH)- ABE	TadA-TadA(7.10)*–XTEN–Sa(KKH)- ABE -Cas9n–NLS	NNNRRT	N9 to N15	N/A	Hua, Tao, & Zhu, 2019 L. Yang et al., 2018

(5). Enhancing base editing with higher efficiency and product purity

To minimize undesired products, which arise from BER by UDG activity, the fourth-generation base editors (CBE4) were engineered (Komor et al., 2016). CBE4 contains two or three copies of UGI to block UDG activity, which leads to significantly lower indel frequencies, higher C to T editing frequencies, compared to CBE3 (Figure 2A) (Komor et al., 2017; L. Wang et al., 2017). Moreover, fusing CBEs to Gam, a bacteriophage Mu protein that binds DSBs, greatly reduces indel formation during base editing. This new CBE can convert target G•C to A•T with high base editing frequencies (~50% in human cells) and very high product purity at very low rates of indel formation (Figure 2A) (Komor et al., 2017).

Furthermore, in order to expand editing window of CBEs, a novel base editing tool was designed, naming it as a base editor for programming a larger C to U (T) scope (BE-PLUS), featuring higher fidelity of base editing and a broader editing window (N5–N17) (Figure 3 and Table 2) (Jiang et al., 2018). Reengineering the sequences of CBEs by codon optimization and incorporation of additional nuclear-localization sequences has enabled target modification in a wide range of mouse, plant and human cell lines (Ren et al., 2017; Zafra et al., 2018). Fluorescence-tracing assay has been established for rapid, efficient and quantitative fluorescent read-outs of DNA editing activity in vivo, and expanded the versatility of CBEs in genome editing and engineering technologies (Shimatani et al., 2017; St Martin et al., 2018). In addition, a series of CRISPR–Cpf1-based CBEs have been developed to overcome the limitation of G/C-rich PAM. The CRISPR–Cpf1-based CBEs recognizes a T-rich PAM sequence and catalyzes C-to-T conversion in the target loci with very low levels of indel formation, and non-C-to-T substitutions (Table 2) (Kleinstiver et al., 2019; X. Li et al., 2018).

FIGURE 3. Schematic view of the base editor for programming larger C to U (T) scope (BE-PLUS).

10 copies of 19-amino-acid GCN4 peptide are fused to the C-terminus of catalytically-dead Cas9 (dCas9) or Cas9 nikase (Cas9n), while Apolipoprotein B mRNA editing catalytic enzyme (APOBEC) and uracil glycosylase inhibitor (UGI) are co-expressed with a single chain variable fragment (scFv) to form a fusion protein. When co-delivery of the two plasmids along with a single guide RNA (sgRNA), dCas9-GCN4 recognizes and binds to the target site, guided by sgRNA, scFv-APOBEC-UGI is recruited around the binding site to induce cytosine (C) to thymine (T) conversion within its editing window.

1.1.2. AID-based DNA base editors

An activation-induced cytidine deaminase (AID), or its orthologue from sea lamprey (PmCDA1) has also been developed (Hess et al., 2016; Ma et al., 2016; Nishida et al., 2016). AID can mediate somatic hypermutation (SHM), which deaminates a cytosine (C) to a uracil (U), initiating a DNA repair response to realize single base mutation (Ma et al., 2016). The targeted AID-mediated mutagenesis (TAM) system combines the AID with dcas9, which can directly convert C or G to the other three bases and generate a large repertoire of variants at desired loci (Table 2) (Ma et al., 2016). This system provides a base resolution and forward genetic tool for screening gain-of-function variants associated with human diseases, and creates new substitutions to enhance protein functionality (Ma et al., 2016). Moreover, CRISPR-X is similar to TAM (Table 2). In order to recruit variant of AID (AID*Δ, the K10E/T82I/E156G AID variant lacking the nuclear export signal sequence), CRISPR-X combines dcas9 with a single guide RNA (sgRNA) bearing two MS2 hairpin-binding sites (Figure 4) (Hess et al., 2016). Mutations with high efficiency (>20%) within the −50bp to 50bp editing window have been achieved (Hess et al., 2016). PmCDA1 linked to the C terminus of Cas9 variants (termed target-AID), and UGI fused to the C terminus of PmCDA1 can suppress indel formation (Table 2) (Nishida et al., 2016). This system can achieve 15%~55% target mutation in mammalian cells (Nishida et al., 2016).

FIGURE 4. Schematic view of CRISPR-X.

Catalytically-dead Cas9 (dCas9) binds the target loci in genomic DNA, which is guided by a single guide RNA (sgRNA) containing two MS2 hairpin-binding sites in its stem loop. Recruiting AIDΔ or AID*Δ through AIDΔ-MS2 or AID*Δ-MS2 fusion protein can induce localized and diverse point mutations.

Notably, fusing AID with ZFN and TALEN can create TALEN-AID and ZFN-AID base editors, which do not rely on PAM (Table 2) (Luhan Yang et al., 2016). After optimizing targeted deaminases, only 13% and 2.5% specific C to T mutation efficiencies have been achieved in Escherichia coli and human cells, respectively. This may be due to inability of TALEN and ZFN to open double-stranded DNA, consequently affecting the efficiency of deaminase (Luhan Yang et al., 2016).

1.2. Creation and evolution of base editor of A•T to G•C in genomic DNA

Because almost half of human pathogenic point mutations come from G•C to A•T transitions, the ability to efficiently convert target A•T base pairs to G•C base pairs can advance the study and treatment of genetic diseases (Figure 5). Thus, the base editing toolbox is expanded by introducing adenine base editors (ABEs) (Gaudelli et al., 2017). Considering E.coli TadA doesn’t require small-molecule activators and can act on polynucleic acid, E. coli TadA was used to evolve a DNA adenine deaminase. TadA was fused to dCas9 to construct a random mutation library. Functional screening was performed using the bacterial chloramphenicol resistance recovery screening system. TadA mutants that efficiently deaminate adenine on DNA were successfully obtained after seven rounds of screening (Figure 6) (Gaudelli et al., 2017). ABEs that fuse TadA mutants with Cas9n (D10A) can effectively convert A•T to G•C on genomic DNA (Figure 1B).

FIGURE 5. Pathogenic human single nucleotide polymorphisms (SNPs) in the ClinVar database.

According to the ClinVar database, there are about 48% pathogenic human SNPs caused by G•C to A•T conversion, and 14% SNPs caused by T•A to C•G conversion. These two types of pathogenic human SNPs can be corrected by adenine base editors (ABEs) and cytosine base editors (CBEs).

FIGURE 6. Scheme of seven generations of adenine base editor (ABE) evolution.

A bacterial selection for base editing by creating defective antibiotic resistance gene that contains point mutations (C•G to T•A mutation) at critical positions. If the mutated ABE (TadA-cas9n-NLS) (nuclear localization signal, NLS) system can convert A•T to G•C at the point mutation, the bacteria will gain resistance and the effective mutated ABE system will be identified. After seven rounds screening and evolution, ABE 7.10 with the highest editing efficiency was obtained. ①: improving editing efficiency by introducing two mutations in TadA*; ②: using homodimer TadA to improve editing efficiency; ③: improving editing efficiency by introducing three mutations in TadA(2.1)*; ④: overcoming sequence preference; ⑤: using heterodimer TadA to improve editing efficiency; ⑥: removing unnecessary mutations by DNA shuffling and improving editing efficiency by introducing one mutation in TadA(5.1)*; ⑦: improving editing efficiency by introducing three mutations in TadA(6.3)*.

: one mutation; : five mutations.

A defective antibiotic resistance gene that contains point mutations (C•G to T•A mutation) was used. If the mutated ABE systems can convert A•T to G•C at the point mutation, the bacteria will gain resistance, and the effective mutate ABE systems will be identified (Gaudelli et al., 2017). The ABE2.1 mutation system with efficiency of 11%±2.9% was created (Figure 6) (Gaudelli et al., 2017). Because TadA natively operates as a homodimer, with one monomer catalyzing deamination, and the other acting as a docking station for the tRNA substrate (Losey, Ruthenburg, & Verdine, 2006), ABE2.9 was later created to offer a higher editing efficiency (20%±3.8%) in the second round screening (Figure 6) (Gaudelli et al., 2017).

Afterwards, ABE5.1 was created, but its editing efficiency was decreased in mammalian cells. To increase the efficiency, ABE5.3 was created by using the wild-type TadA instead of evolved TadA* variant in the N-terminal TadA domain, and the efficiency was increased up to 39±5.9% (Figure 6) (Gaudelli et al., 2017). Finally, ABE7.10 was screened to convert target A•T to G•C base pairs with efficiency of ~50% in human cells at a very high product purity (typically ≥ 99.9%), and very low rates of indels (Gaudelli et al., 2017).

Based on evolutionary of CBE, ABE was also developed to expand the editing scope by using natural and engineered Cas9 variants with PAM specificities (Table 2) (Hu et al., 2018; Hua, Tao, & Zhu, 2019; L. Yang et al., 2018). Studies have shown that expression levels of base editors are major bottlenecks for base editing efficiency. CBE4max, AncBE4max and ABEmax editors have been developed by adopting bipartite nuclear localization signals (bpNLS), to achieve increased editing efficiencies in a variety of settings, especially under suboptimal conditions or at sites previously edited with low efficiencies (Koblan et al., 2018).

2. Applications of CRISPR-mediated base editing

Precise editing is crucial for successful biomedical research and gene therapy. The newly developed CBE and ABE systems have provided an efficient tool for precise genetic modification of the murine and human genome.

2.1. Exerting base editors in gene knockout

Based on the high efficient and precise editing of the CBE systems, which can fix point mutations of C to T or G to A in its mutation window, it is convenient to change the genetic codon (missense mutation) in an open reading box or produce the termination codon in advance. Gene knockout can be achieved if any of the four codons (CAA, CAG, CGA, and TGG) are converted into the gene termination codons (TAA, TGA and TAG) (Billon et al., 2017). Several studies have used base editing systems to inactivate genes by altering genetic code to create stop codons (named CRISPR-STOP and iSTOP method) (Billon et al., 2017; Jia et al., 2019; Kuscu et al., 2017; Zhang et al., 2018). Wild type Cas9 can lead to more DNA damage and cell death when targeting a gene. Notably, the Cas9 enzyme will produce multiple DSBs, therefore, causing chromosomal structural abnormality when targeting high-copy-number genomic regions. The CRISPR-STOP approach is a less deleterious and more efficient gene silencing alternative to wild type Cas9, and it is also expected to be applicable to genome-wide functional screenings (Kuscu et al., 2017).

The CBEs can introduce precise base conversion without causing double-chain fractures, while avoiding extra mutations such as target and non-target sequence deletion and insertion caused by DSBs. For instance, proprotein convertase subtilisin/ kexin type 9 (PCSK9) can promote degradation of low-density lipoprotein receptor (LDLR), and increases the level of low-density lipoprotein cholesterol (LDL-C), which is an important target for lowering cholesterol level and cardiovascular diseases. It has been reported that knockout of PCSK9 in the liver by the traditional CRISPR-Cas9 system can significantly reduce the level of LDL-C in the blood (Ding et al., 2014). However, a recent study has shown that using CBEs to knockout the PCSK9 gene in the mouse liver leads to a substantial decrease in plasma PCSK9 protein levels (>50%), and cholesterol levels in plasma (about 30%), and causes no off-target mutagenesis, neither cytosine-to-thymine edits nor indels (Chadwick, Wang, & Musunuru, 2017). Moreover, loss-of-function mutations in angiopoietin-like 3 (ANGPTL3), created by base editors, have provided a prospective strategy to treat patients with atherogenic dyslipidemia (Chadwick, Evitt, Lv, & Musunuru, 2018).

Altogether, base editing systems have offered advantages over traditional HDR-based CRISPR-Cas9 genome-editing methods, and have been demonstrated as a robust and efficient gene disruption technology compatible with genome-wide studies to investigate gene functions, and to model human diseases.

2.2. CRISPR-mediated base editing in mice

The CBE- and ABE-mediated systems have provided a new and efficient tool for single-nucleotide modification of the mouse genome. The CBE or ABE systems have been used to create single-nucleotide variation in mouse models (K. Kim et al., 2017; Liang, Sun, et al., 2017; Liang et al., 2018; Liu et al., 2018; Ma et al., 2018; Ryu et al., 2018; Yeh, Chiang, Rees, Edge, & Liu, 2018).

Delivering CBE3 ribonucleoproteins (RNPs), which target the Dmd or Tyr gene via electroporation or microinjection into mouse zygotes, led to efficient and precise base editing (K. Kim et al., 2017). Studies have shown that the CBE system is more efficient than the previously used TALEN, Cas9 and Cpf1 nuclease for site-specific mutagenesis in mice (Hur et al., 2016; K. Kim et al., 2017; Sung et al., 2013; Sung et al., 2014). When HF2-BE2 or ABE was introduced into mouse zygotes, an efficient base editing was observed in both embryos and live born mice (Liang, Sun, et al., 2017; Liang et al., 2018). Moreover, the ABE system has been shown to possess efficient and precise base editing in rats (Ma et al., 2018; Ryu et al., 2018).

Utilizing the ABE and CBE systems, clinically relevant mutations in androgen receptor (Ar) and homeobox D13 (Hoxd13) genes, and multiple mutations have been created in mouse models (Liu et al., 2018). The Dmd nonsense mutation in a mouse model of Duchenne muscular dystrophy has been corrected by using the ABE7.10 system, demonstrating a therapeutic potential of base editing in adult animals (Ryu et al., 2018). Combining the base editing system with semi-cloning technology to screen pivotal amino acids for Dnd1 gene, mutations that cause mouse primordial germ cell deficiency and infertility have been identified (Q. Li et al., 2018; Youngren et al., 2005). This novel strategy has provided an effective tool for in vivo screening of amino acids that are crucial for protein function at the organismal level.

A challenge in genome engineering is to simultaneously introduce mutations into linked loci (located on the same chromosome). Introducing C•G-to-T•A transition into two cytokine-sensing transcription factor binding sites separated by 9 kb using base editing has revealed that one enhancer activates two flanking genes in mammary tissues during pregnancy and lactation (H. K. Lee et al., 2019). Thus, introducing linked mutations simultaneously in one step can help understand linked cis-elements related to disease models and pathogenic mutations (H. K. Lee et al., 2019).

2.3. CRISPR-mediated base editing in human cells

The single-base editing technology has been used not only in mice to model human diseases, but also directly in human cells (G. Li et al., 2017; Liang, Ding, et al., 2017; Yuan et al., 2018; Zeng et al., 2018; Zhou et al., 2017). The CBE3 and SaKKH-BE3 (PAM is NNNRRT) systems have been used to mutate one or three genes simultaneously in human trigeminal zygotes with very high efficiency, indicating that CBE3 induces near perfect gene editing in the target site with extremely low off-target mutagenesis and indel mutations in human cells (G. Li et al., 2017; Zhou et al., 2017).

The base-editing strategy has been used to correct disease mutants in human cells, for instance the beta-thalassemia and marfan syndrome pathogenic FBN1 mutations (Liang, Ding, et al., 2017; Zeng et al., 2018). These studies have demonstrated the feasibility of curing genetic diseases in human cells using the base editor system.

Mutations in the isocitrate dehydrogenase 1 (IDH1) gene are associated with a number of cancers such as gliomas (Hartmann et al., 2009; Kang et al., 2009; Mardis et al., 2009; Yan et al., 2009). Establishing sustainable cellular models harboring IDH1 mutations is difficult (Piaskowski et al., 2011). The heterozygous IDH1 R132H mutation (IDH1^R132H/WT) in human astroglial cells has been successfully created as a sustainable mutated IDH1 model using the base-editing strategy (Wei et al., 2018).

RNA splicing is a critical mechanism by which to modify transcriptome, and its dysregulation is associated with many human diseases. Intriguingly, a report has shown that TAM can be used to modulate RNA splicing by editing splice sites (Yuan et al., 2018). Thus, the CRISPR-guided cytidine deaminase provides a versatile genetic platform to modulate RNA splicing and correct mutations associated with aberrant splicing in human diseases (Yuan et al., 2018).

3. Challenges of base editors

3.1. Target limitations

The target site is limited by both PAM and editing window. PAM is required for targeting a DNA site by CRISPR family nucleases, and it should be appropriately positioned relative to the target base to ensure efficient editing. Thus, the use of base editors is still limited, even though natural and engineered Cas9 variants with different PAM specificities have been developed (Hu et al., 2018; Hua et al., 2019; Y. B. Kim et al., 2017; Koblan et al., 2018; L. Yang et al., 2018). Moreover, the deamination window of base editors mainly depends on which deaminase is fused. Although base editors using mutated deaminase domains can narrow the width of the editing window from ~5 nucleotides to as little as 1–2 nucleotides, its efficiency is lower than the natural cytidine deaminase (Y. B. Kim et al., 2017). There are still no ideal base editors can precisely target one base with high efficiency, which has limited application of base editors in gene therapy, in particular, pathogenic single nucleotide polymorphisms (SNPs).

Some researchers have tried to combine deaminase with other programmable nucleases such TALENs or ZFNs to eliminate PAM restrictions and achieve full genome coverage. However, the result is very inefficient compared to the base editor system (Luhan Yang et al., 2016). The reason might be that deaminase mainly acts on single-strand DNA, and TALENs or ZFNs can’t create single-strand DNA “bubble”, which may reduce the catalytic efficiency of deaminase.

3.2. Indels and off-target editing with DNA base editors

Although single-base editing systems do not induce DSBs, little insertion or deletion (indels) still exist in the target locus. Studies have shown a 20 base pairs (bps) deletion when editing in mouse embryos (K. Kim et al., 2017; Zafra et al., 2018; Zong et al., 2017). It is likely that these indels are caused by Cas9n, nicking the non-target strand (K. Kim et al., 2017). Even though HF2-BE2 has been used to avoid single-strand breaks, indels still exist in target locus (Liang, Sun, et al., 2017). However, when UGI is incorporated into base editing systems, indels are substantially reduced (Komor et al., 2016).

Similar to the traditional CRISPER-Cas9 technology, both cytosine and adenine DNA base editors have the potential to induce off-target editing. Off-target base editing can be classified into “off-target editing within the editing window”, which is inevitable off-target if the non-target C or A in the editing window; “proximal off-target editing”, which is the editing that takes place outside the editing window, but near the target locus, for example within 200 bp; and “distal off-target editing”, which is the editing that occurs away from the target locus (D. Kim et al., 2017).

A study has used digested-genome sequencing (Digenome-seq) to assess specificity of the BE3 editing system and the Cas9 enzyme, which target EMX1 and HBB genes in vitro. Although the off-target frequency of the BE3 system is far less than the Cas9 enzyme, off-target sites were still found using the BE3 system (D. Kim et al., 2017). A substantial reduction of off-target editing has been achieved when Cas9 high-fidelity variants, which contain specific point mutations, have been used to generate high-fidelity versions of base editors (J. K. Lee et al., 2018). Moreover, a study has shown that the CBE with rAPOBEC1 and ABEs can cause extensive transcriptome-wide RNA cytosine deamination in human cells (Grunewald et al., 2019). CBE-induced RNA editing occurs in both protein-coding and non-protein-coding sequences and generates mutations of missense, nonsense, splice site, 5’-untranslated region (UTR), and 3’-UTR. Two CBE variants containing rAPOBEC1 mutations can substantially decrease the numbers of RNA editing in human cells. These variants also show more accurate on-target DNA editing (Grunewald et al., 2019).

Taking together, scientists need to fully define and characterize DNA and RNA off-target effects of deaminase enzymes in base editor platforms in order to ensure safety of gene therapy. Higher fidelity versions of base editors, or base editor variants that do not rely on the CRISPR-Cas9 system need to be developed to minimize random non-directed off-target base editing and indels.

3.3. Base substitution limitation

There are some known pathogenic SNPs caused by other types of base-pairs mutations such as C•G to A•T, A•T to C•G, C•G to G•C, G•C to C•G, A•T to T•A, and T•A to A•T, according to the ClinVar database (Figure 5). However, the CBE and ABE system now can only achieve A•T to G•C and G•C to A•T substitution in genomic DNA, which limits its application (Gaudelli et al., 2017; Komor et al., 2016). Thus, it is timely to customize new enzymes to expand the application of base-editing system.

4. Prospects of base editing

CBEs have been proven to be the new favorite genome editing systems, due to their advantage of being efficient, precise and irreversible base editing. Together with the development of ABEs, application of base editors has been extensively expanded in base-pairs substitutions of G•C to A•T, and A•T to G•C in genomic DNA. CBE-systems have been used to silence genes by introducing stop codons, to correct SNPs (Billon et al., 2017). Moreover, they provide a new genetic tool to screen gain-of-function variants at base resolution (Ma et al., 2016). Thus, base editors will emerge as a powerful tool in functional screening. Even though limited editing window size, off-target effects and cytotoxicity are still some problems to be solved, base editors are becoming a magical tool in single-nucleotide editing (D. Kim et al., 2017). Single-base editing technology has shown great promise for applications in an array of species, and is a highly favorable tool in biomedical research and gene therapy.